How a 3-person store answers support at 2am (without hiring)

Somewhere around thirty orders a day, every small store discovers the same ugly ratio: each order generates about a third of a support message. "Where is my order" alone will be half your inbox. Behind it, in reliable proportions: refund and return questions, size exchanges, COD confirmations, "I ordered the wrong color," and the occasional payment that failed halfway and left money in limbo.

Thirty orders a day is 10–15 conversations a day. Not enough to justify a hire. Far too many to ignore — because an ignored WISMO message doesn't stay a status question. Left unanswered for 48 hours it matures into a refund demand, a chargeback, or a one-star review, each of which costs more than the answer would have. And these messages don't arrive during business hours. They arrive when your customer remembers the parcel exists — in bed, at 2am, three days after the courier picked it up.

A three-person store has three options. Answer everything yourselves and slowly lose your evenings. Hire before the revenue justifies it. Or make most of the answers send themselves. This is the playbook for option three.

The loop, in four steps

The mechanism is a loop, not a bot-with-a-dashboard:

1. Write down the answers you already give, as short help center articles.
2. The AI answers customers from those articles — and only those articles. It doesn't improvise your return window from what stores "usually" do.
3. When it doesn't know, it stays silent and the conversation simply waits for a human, like it always did. Every silent miss is logged as a question, deduplicated, and counted.
4. The most-asked misses become next week's articles. Coverage grows exactly where your customers push on it.

Step 3 is the part that makes this trustworthy enough to leave running overnight. An AI agent that answers everything will eventually invent a policy at 2am and you'll wake up to the fallout. One that answers only what you've published, and hands the rest to a human, can't embarrass you — the worst case is the status quo: a question waits until morning.

Seed it in one evening: the first 15 articles

Don't aim for complete documentation. Aim for the fifteen answers you're sick of typing:

• Delivery (5): How long does shipping take (by region, honestly)? How do I track my order? What if tracking hasn't moved in three days? Do you ship internationally? What are the shipping charges?
• Returns & refunds (4): What's the return window and what condition counts? How do I start a return? When does the refund arrive after you receive it, and to where? What's exchangeable vs. refund-only?
• Payments & COD (3): Is COD available and is there a fee? Payment failed but money was deducted — what now, and how long does the auto-reversal take? What payment methods do you accept?
• Orders (3): Can I change my address after ordering? Can I cancel, and until when? Can I change the size or color before dispatch?

Each one is 100–250 words in the help center editor. Your source material is your own sent-messages history — you've written all of these before, many times, just never in a place that remembers. Publishing an article makes it part of what the AI can answer, immediately.

What 2am actually looks like

The WISMO message: "ordered 5 days ago where is my order???" The AI answers with your actual dispatch and delivery timelines and how to use the tracking link — from your article, in seconds, at 2:14am. Most WISMO askers don't need intervention; they need a confident restatement of the timeline from someone (something) that sounds sure. By morning the thread is resolved and never touched your queue.

The refund policy question: "what if the kurta doesn't fit?" Policy is knowledge — the AI answers the window, condition, and process. But an actual "please refund order #4312" is money, not knowledge, and money keeps a human in the loop: the thread routes to your morning queue instead. The AI answers about refunds; it doesn't grant them. That line is where customer trust — and your accounting — lives.

The COD hesitation: "is COD available in Indore? extra charge?" Answered from your COD article instantly. This one matters more than it looks: COD questions are usually pre-purchase, and an unanswered pre-purchase question at 2am is an abandoned cart by 2:05.

The gaps list writes your roadmap

Open Insights once a week and read the knowledge gaps — every question the AI stayed silent on, ranked by how many customers asked it. The first week is humbling: questions you never thought to document ("do you restock sold-out sizes?", "is there a bulk discount for 10+ pieces?"). Each row has a write-article button that starts the draft with the customer's own phrasing as the title.

Two or three articles a week and the compounding is fast: stores that start with 15 articles are typically past 40 within a month — and each new article closes that question for every future customer at once. As a side effect, the help center is a public site on your own domain, so "yourstore return policy" starts resolving on Google to your page instead of nowhere.

The math

Take 20 conversations a day. With seeded coverage of the repetitive majority — and WISMO, returns, COD, and shipping are the majority — 60% resolving automatically is a normal outcome once the gaps list has been fed for a few weeks. That's 12 conversations a day you don't handle, at four-ish minutes each: about 50 minutes a day, 25 hours a month — most of a working week, every month, for a team of three.

And that undercounts it, because it prices a WISMO answer at four minutes when its real value is the chargeback it prevented and the cart it saved. Faster answers aren't just cheaper support; they're revenue defense.

Setting it up

The whole thing is an evening plus a toggle: create a workspace at signup (a form, not a sales call), paste the two-line snippet on your store, write the fifteen articles, turn the AI on. Keep your mornings for the queue that's left — the genuinely human threads — and your evenings for yourself.

You'll still do support. You'll just stop doing the same twelve answers of it, forever, at 2am.